[OpenIndiana-discuss] sata hba choice

[Sašo Kiselkov]

On 08/26/2012 01:10 AM, Timothy Coalson wrote:
> On Sat, Aug 25, 2012 at 1:56 PM, Sašo Kiselkov <skiselkov.ml at gmail.com> wrote:
>> On 08/25/2012 03:42 PM, Jordan McQuown wrote:
>>> I can say from experience chaining supermicro sas chassis with any type of disk can be a recipe for disaster. Lsiutil reported a couple million errors in a couple months time which eventually led to some sort of contention and corruption on both sides of our mirrors. These were also SAS and not sata disks.
>>
>> Have you checked your SAS cabling? Also, lsiutil is for lsi1068 (3G) SAS
>> chips. I've got a SuperMicro SC847 attached to a pair of lsi
>> sas2008-based boards. The chassis is internally arranged as a pair of
>> backplanes, each with two LSI SAS2x36, chained to appear as a single
>> connector from the back of the chassis (i.e. one connector per SAS
>> channel bundle for two fully redundant paths to the drives). Initially I
>> was having trouble with excessive link errors which resulted in poor
>> performance, lots of errors and generally lots of problems, but I found
>> this was caused by a faulty SAS cable between the host and the chassis.
>> Ever since playing around with cabling a bit it's been ticking reliably
>> like a Swiss clock.
>>
> 
> Good to know it is working for you, that is precisely the kind of
> setup I was considering.  Did I understand you correctly that even
> with redundant paths to each drive, a single poor connection, despite
> a secondary good connection, caused serious performance problems?
> 
> Thanks for the replies.

Correct. The problem was that the scsi_vhci driver was trying a
round-robin approach to load-balancing, so half the commands were
hitting the good path and half the bad one (which lagged and lead to
losses). There is no redundancy mechanism that is impervious to
non-gracious hardware failures, and storage systems are commonly
affected by this (despite what any manufacturer tells you).

--
Saso